Research On Super-resolution Imaging Techniques Of Multi-frameaerial Images

With the rapid development of the aerial camera,we want to achieve the larger image,higher image resolution and the farther distance image.But affected by volume,weight and power of the aerial camera and other factors,such as under-sampling,motion blur and noise caused by the optical imaging process,the aerial image resolution can not meet the application requirements.Therefore obtaining higher resolution aerial images has become a hot and difficult point in the field of aviation.The most direct method of improving the image resolution is using high-resolution CCD camera.But due to the limitation of the technology level and the speed of the camera image data transmission,the ability of high-resolution image obtained by high-resolution CCD camera is very limited.In recent years,through the signal processing method to improve the image resolution-super resolution technology is widely concerned.On the basis of not changing the original hardware imaging system,super resolution technology uses the signal processing method to make one or several low resolution images which contain similar information and different details to form a high-resolution image."Super" represents the purpose of overcoming the inherent diffraction limit of traditional low-resolution imaging system.It can get the spatial frequency information beyond the limit and achieve the purpose of further improving the resolution.This dissertation firstly discusses the basic theory and the engineering application of dynamic image super resolution reconstruction technology.Then this paper summarizes,analyzes and compares the physical imaging models,methods and evaluation systems.Based on imaging model,we have conducted in-depth research on the topic-multiple image superresolution technology to resolve the computational complexity,edge blur and image distortion of the existing super-resolution algorithms.The main innovations and research results of this dissertation are as follows:1.In order to enlarge a low resolution image clearly,an improved Papoulis-Gerchberg super-resolution method was proposed to solve the space complexity and the edge blurring phenomenon of reconstruction results.The algorithm uses edge detection operator.And canny detection is joined in every Papoulis-Gerchberg iterative process,while reconstruction error is projected to nextiterative process.So that new algorithm reduces the space complexity and recovers the lost the high-frequency edge information effectively.Compared with other conventional super-resolution method,the proposed algorithm can reconstruct multi-frame Low-Resolution images of same scene more accurately and the visual quality of the reconstruction image is clearer.Meanwhile,the proposed algorithm can eliminate edge shadow and obtain a clear high-resolution image.2,In this paper,a super-resolution technology of combining hardware and software is studied.Firstly,multiple images of the same scene produced by different motion parameters are chosen to be both training set and input images.Secondly,regarding the fact that traditional LLE super-resolution technology over-relying on external training images,we propose a self-learning algorithm based on the local linear embedding(LLE).The new LLE weight calculation method is proposed to get initial estimate of HR image.Meanwhile,we use self-learning LLE algorithm to recover lost high-frequency information of initial estimate and obtain the final estimate.This algorithm can provide high quality reconstruction image,improve the resolution of the captured image effectively and satisfy the demand of gaining high quality image.3.Despite that general Dictionary-Based Super-Resolution algorithms obtain redundant dictionaries from numerous HR-LR images,HR image distortion is unavoidable with the absence of the input images' frequency information in the corresponding HR-LR images.To solve this problem,this paper proposes a multi-frame super-resolution reconstruction algorithm based on self-learning methods.First,multiple images from a same scene are selected as both input and training images,and larger scale images,which are also involved in the training set,are constructed from the learning dictionary.Then,different larger-scale images are constructed via repetition of the first step and the initial HR set whose scale closely approximates that of the target HR image are finally obtained.Lastly,initial HR images are fused into one target HR image under the NLM idea,while the IBP idea is adopted in order to meet the global constraint in the reconstruction process.The simulation results demonstrate that the proposed algorithm produces more accurate reconstructions than those produced by other general super-resolution algorithms.4.This paper introduces a new high-performance super-resolution(SR)method for multi-frame images.By combining learning-based and reconstruction-based SR methods,this paper proposes a multi-frame image super-resolution method based on adaptive self-learning.Using the adaptive self-learning method and recovery of high-frequency edge information,an initial high-resolution(HR)image containing effective texture information is obtained.The edge smoothness prior is then used to satisfy the global reconstruction constraint and enhance the quality of the HR image.Our results solve the distorted disadvantages of reconstruction-based and learning-based methods and achieve better performance than several other methods.It is beneficial to improve the resolution of aerial image.In conclusion,according to a series of problems such as fuzzy distortion in all kinds of super-resolution algorithms,this paper analyzes many related techniques involving multiple image super resolution reconstruction.And it has gotten primary achievements.All these researches not only provide a theoretical basis for the further research,but also improve the spatial resolution and the detail definition of aerial image.